Download The ground plane: Lord of the Board

The ground plane: Lord of the Board
The ground plane: Lord of the Board
Tim Williams
Elmac Services
www.elmac.co.uk
Introduction
The way that a circuit is laid out on a PC board is critical to its EMC performance. At
high frequency all tracks must be regarded as components in their own right, since the
coupling of high frequency signals from one part of the circuit to another is mediated by
their stray capacitance and inductance. Even if the equipment is a simple analogue circuit
with no HF emissions sources, susceptibility to RF disturbances means that it must still
be laid out with RF performance in mind. The schematic rarely if ever tells this side of
the story; PC tracks appear as nothing more than lines on the diagram joining pins of
lumped components such as ICs, resistors and capacitors.
But every track on a board exhibits its own partial self inductance, and mutual inductance
and capacitance to other nearby areas of copper, and if these are uncontrolled then the
performance of the board at high frequency – and that means the EMC performance of
the circuit – is also uncontrolled. The simplest and most effective way of fixing the
characteristics of all such tracks is to implement a ground plane on the board.
How a ground plane works
In this article, the term "ground plane" will be synonymous with "0V plane". That is, the
reference (0V) for the power supply will be taken to be the same as "ground". This is not
necessarily the same as "chassis ground" and there is no general rule (although there is a
preference) as to whether the chassis should be connected to 0V or not. Since we are
discussing just the PCB layout here, the issue will be largely avoided.
First of all, forget any notion of the ground plane being used for screening: it may do, but
that is not its primary function. What you expect from it is to provide a high-frequency
return path for every signal trace and power rail on the board.
Running a signal and its ground return close together – so that neighbouring currents
flow in opposite directions – reduces the inductive impedance of the total path by a factor
equivalent to the mutual inductance between the two halves:
Effective inductance of signal loop
=
2(L - M)
where L is the self-inductance of each half of the loop, M is the mutual
inductance between the halves
Page 1
The ground plane: Lord of the Board
Bad practice: wide separation of signal and return, hence low M
Good practice: close coupling of signal and return, M ≈ L
Figure 1 Separation of signal and return
This effect, which "kicks in" above a few tens to hundreds of kHz, is critical to
understanding PCB track routing at radio frequencies: you need to control the 0V current
return path. Keeping the signal and 0V paths adjacent along their length – and therefore
maximizing their mutual inductance – will ensure not only minimum coupling with the
magnetic fields around the PCB, but also minimum impedance of the return path and
therefore minimum noise voltage developed along it. This will keep down the “noisiness”
of the whole board, which is mostly what is responsible for common mode RF emissions.
It will also improve its susceptibility to external interference, which is created by the
conversion of incoming common mode disturbances to differential mode within the
circuit.
Now, it's entirely possible to do this by assigning a 0V return track to each signal and
running the two next to each other, or on adjacent layers, throughout the board. This is of
course somewhat tiresome, not to say awkward, in a board with hundreds of such paths,
and it isn't necessary. If you provide a 0V (ground) plane as a continuous layer adjacent
to each track layer, then the job is done for you. The return current for each high
frequency signal track automatically finds its own preferential path through the plane,
which will be that path which is directly underneath the signal: the geometry ensures that
this path has the least enclosed loop area and therefore the least total path inductance. All
you have to do is make sure that the component 0V connections, which carry the return
path through, have a short and direct connection to the 0V plane, which is easy; and that
the 0V plane remains unbroken along the length of any of these paths, which despite also
being easy on a multi-layer board still occasionally suffers from poor implementation.
Ground planes in the xy direction: breaks, moats and joins
The effect of a slot across the current flow is to divert the return current from its optimum
path and increase the 0V return path inductance. There are two EMC problems caused by
extra inductance here:
Page 2
The ground plane: Lord of the Board
•
one is that any high frequency current passing through this inductance will develop a
voltage across it. This voltage then appears across the two parts of the plane, each of
which has a high capacitance to the environment due to its surface area. On its own
this structure forms an efficient radiating dipole, which is then enhanced if cables are
attached to the board;
•
the other is that the inductance itself represents increased magnetic coupling to the
environment through the enclosed loop area.
Therefore either there should be no split within the plane to add this extra inductance, or
high frequency currents should be prevented from flowing between the two separated
segments. This can be achieved through careful layout design for emissions control, by
not running any tracks carrying such current across the gap. But it doesn’t address the
problem of immunity to incoming RF fields. The magnetic and electric coupling is still
present for these fields, and so they will develop voltages between the two ground
segments which can create circuit susceptibilities.
Figure 2 The effect of a discontinuity on return current
No split: high mutual inductance,
low ground inductance
Iret
Iret
Extra return path adds ground inductance
Iret
Iret
VN
generates noise across radiating dipole
There is a convention, encouraged by many device application notes, to create different
0V planes for different parts of the circuit: for instance, analogue 0V and digital 0V,
which are then linked at one point on the board, typically at an A-D converter. The
purpose of this trick is to prevent digital noise currents from flowing in the analogue
circuit and corrupting it, that is, it relates to internal EMC. Unfortunately it creates
problems for external EMC, and introduces difficult questions that could be avoided by a
single system-wide 0V plane:
•
where should you join the planes when there are multiple analogue-digital interfaces?
•
what about power returns that are common to both analogue and digital?
Page 3
The ground plane: Lord of the Board
•
where do you run signals that must cross the break between the analogue and digital
sections? Any of these signals will be exposed to incoming interference that will
appear between the planes, or will develop such interference themselves.
•
if you are going to bond the 0V planes at RF to the chassis of the product, which is
always a recommended practice to minimise common mode coupling into or out of
the PCB, how do you choose which of the planes should be bonded?
single point connection at ADC
ADC
analogue & digital currents
kept separate, but...
...no local return path for
other signal tracks unless
carried across the link,
so large loop area
Any split ground
plane is an effective
dipole structure
can be grounded to
chassis
cannot be grounded to
chassis (or vice versa)
Figure 3 The problems of split ground planes
For all these reasons, a preferable way to proceed is to use a single, system-wide 0V
plane and carefully lay out the board so that digital noise is effectively segregated from
low-level analogue circuits. Also, lay out the analogue circuits so that the critical nodes
are subject as little as possible to ground noise voltages, that is, keep sensitive signals
free of common ground impedance. Avoid as far as possible any discontinuities in this
plane and make sure that you do not run any critical (noisy or low-level) tracks across
any discontinuities that remain, such as slots formed by close-packed vias, or near to the
edges of the plane. There is no merit in putting "moats" in this 0V plane.
In the centre of a plane inductive effects are minimal and the plane impedance is
dominated by resistance, but towards the edges fringing magnetic fields are created and
the inductive impedance of the plane rises. A good rule of thumb is to keep critical tracks
further away than 10·h from the edge of the plane or the edge of any discontinuity, where
h is the separation distance in the z-direction between the layer that carries the track and
the layer that carries the plane.
This can also apply to power plane layers placed against a 0V plane. If the power and 0V
planes were to have exactly co-located edges, the transmission line formed by the two
planes would be terminated in a sharp open circuit and the problem of transmission line
resonances would be potentially severe, leading to a high degree of coupling to and from
Page 4
The ground plane: Lord of the Board
the PCB at the resonant frequencies. Staggering the edges of the planes softens the
impedance mismatch and reduces the Q of the resonance. In fact, in contrast to the
recommendation not to split the 0V plane, it is a good idea to split power planes into
several smaller planes each of which supplies a particular segment of the circuit, and to
decouple these from each other with small ferrite chip inductors.
The principle exception to the "no slots or moats" rule is where there is a good reason to
avoid particular circuits coupling to the main 0V. Almost always, this is when such
circuits are galvanically isolated: interface circuits isolated using opto-couplers, for
instance, or data circuits using transformer isolation such as Ethernet. In these cases,
there should be a window in the main ground plane within which all such isolated
circuitry should sit. The edge of the 0V/ground plane should come up to halfway across
the isolation barrier. Such isolated circuits need to be laid out as a unit so that there is no
compromise of the isolation barrier, or the edge of the ground plane, by other circuits.
separate isolated
ground plane
processor
section
ground plane
window at Ethernet
isolation boundary
power
opto-isolators
circulating currents kept
away from sensitive
analogue section
transformer
Ethernet
isolated section
isolation barrier
analogue section
no tracks crossing the split
Figure 4 Uses of ground plane windows
Simpler circuits may use double-sided boards on which one side is devoted to a ground
plane and the other side to all tracking, and the components. A good, solid plane can be
implemented with some effort, but almost invariably some tracking has to be carried on
the plane side, resulting in some apertures in the plane. The rule here is emphatically not
to split the plane more than is absolutely necessary: keep any such tracks as short as
possible and only use them as links, not to travel long distances. The resulting geometry
of the cut plane should be analysed to check the effect of the cuts on return current flows
under critical tracks.
Ground planes in the z direction: layer stack-up
Successful implementation of a 0V plane on a PCB requires careful attention to the layer
stack-up, that is, the order in which the different tracking purposes are assigned to
different layers. In multilayer (4 layers or more) configurations, every signal layer should
be adjacent to a 0V or power plane layer. Also, power and 0V planes should be on
Page 5
The ground plane: Lord of the Board
adjacent layers, possibly with deliberately thin separation, to take advantage of the
interlayer capacitance for high frequency decoupling. Critical tracks, for instance those
carrying high di/dt signals such as clocks, should be routed adjacent to 0V rather than
power planes. Such tracks should also not jump through vias from one 0V plane layer to
another, unless the 0V layers are tied together with vias near that point.
For boards with more than 6 layers, multiple 0V plane layers are needed so that tracks
and power plane layers can always be located next to a 0V plane, but it is also essential to
tie these together with grounding vias at small intervals – intended connections to 0V
will provide this automatically, but in areas of the board where this doesn’t happen, a
1cm spacing of such “stitching” vias will assure good performance of the planes up to
about 2GHz.
Some consideration needs to be given to the question of whether 0V planes on different
layers should or shouldn't overlap. Generally, a common geometry on all layers is easiest
to work with and avoids the question. If there is a need for some reason to have partial
overlaps, the principal consequence is that the inter-plane capacitance may create
unexpected high frequency resonances, and the more complicated the structure, the
harder it becomes to anticipate and control such effects. But don't assume they are
negligible: as a simple example, a 2cm2 overlap with a separation distance of 0.3mm in
FR4 fibreglass laminate will have a capacitance of about 26pF; with 2nH of inductance
in parallel (possible, with lack of attention to other details) this will resonate at 698MHz,
quite capable of causing trouble in EMC tests.
One consequence of the dependence on layer separation is that the detail of the vertical
sandwich construction of the PCB can be as important for its EMC as is the layout
pattern. The separation is determined by the thickness of the pre-preg and core materials
used in the build of the bare board. The danger is that this is left to the PCB manufacturer
and is not specified by the circuit designer, or anyone else. As a result you end up with
the default thicknesses used by a particular board supplier, which might be perfectly
adequate for the EMC performance of the product and so is never questioned; but if
another board supplier is chosen during the product life cycle, it is entirely possible that a
different set of thicknesses could be used which result in changed performance. To guard
against this, make sure that you specify layer thicknesses in the PCB drawing, if
necessary checking with your preferred supplier what their defaults will be.
Constant impedance tracks over the ground plane
It is commonplace to find that some tracks, because they carry very high frequencies or
are routed long distances, have to be treated as transmission lines so that reflections along
their length are minimised. This contributes both to good external EMC in the form of
lower emissions, and to good signal integrity, improving the reliability of the circuit. This
means that the tracks have to avoid discontinuities in their linear and cross-sectional
geometry. This in turn requires that
•
they should only run in one layer, with no vias except at their ends
•
this layer should be adjacent to one (microstrip) or two (stripline) continuous 0V
planes
•
the track cross-section dimensions and dielectric constant of the layer should be
defined and controlled throughout the layer to give the correct constant transmission
Page 6
The ground plane: Lord of the Board
line impedance: in practice, impedances between 50 and 100 ohms are easiest to
achieve
•
changes in direction of these tracks should be gradual rather than abrupt
This then requires that at least one layer on a PCB is devoted to these "constant
impedance" tracks, and at least one adjacent ground plane layer (completely unbroken) is
essential to maintain their characteristics.
Summary
The ground plane has enough functions on any electronic circuit board to ensure that at
least one layer needs to be devoted to it alone. For all multilayer boards, it can be
incorporated as a matter of course. Even for low cost double sided boards, with care a
ground plane can be implemented on one side, provided that the rules about avoiding or
dealing with breaks are understood. When such a plane is implemented properly, the
external EMC of the board is automatically improved. Even simple analogue circuits are
susceptible to exposure from radiated RF fields in the environment, and any designer
who consciously avoids putting in a ground plane, on the grounds of cost, will find that
satisfactory EMC performance will need greater cost and effort than if the plane had been
implemented from the start.
Tim Williams is a consultant with Elmac Services and is the author of EMC for Product
Designers (4th Edition, Elsevier 2006)
Copies of this paper are freely downloadable from www.elmac.co.uk
Originally published in EMC Journal Issue 72 September 2007
Elmac Services, Wareham, Dorset: phone 01929 558279
Page 7